Lead Research Organisation: University of Liverpool
Department Name: Mech, Materials & Aerospace Engineering


Nuclear power has great potential as a future global power source with a small carbon footprint. To realise this potential, safety (and also the public perception of safety) is of the utmost importance, and both existing and new design nuclear power plants strive to improve safety, maintain availability and reduce the cost of operation and maintenance. Moreover, plant life extensions and power updates push the demand for the new tools for diagnosing and prognosing the health of nuclear power plants. Monitoring the status of plants by diverse means has become a norm. Current approaches for diagnosis and prognosis, which rely heavily on operator judgement on the basis of online monitoring of key variables, are not always reliable. This project will bring together three UK Universities and an Indian nuclear power plant to directly address the modelling, validation and verification changes in developing online monitoring tools for nuclear power plant.
The project will use artificial intelligence tools, where mathematical algorithms that emulate biological intelligence are used to solve difficult modelling, decision making and classification problems. This will involve optimizing the number of inputs to the models, finding the minimum data requirement for accurate prediction of possible untoward events, and designing experiments to maximize the information content of the data. We will then use the optimised system to predict potential loss of coolant accidents and pinpoint their specific locations, after which we will progress to prediction of possible radioactive release for various accident scenarios, and, in order to facilitate emergency preparedness, the post release phase will be modelled to predict the dispersion pattern for the scenarios under consideration. Finally, all of the models will be validated, verified and integrated into a tool that can be used to monitor and act as an early warning device to prevent such scenarios from occurring.


10 25 50
publication icon
George-Williams H (2018) Extending the survival signature paradigm to complex systems with non-repairable dependent failures in Proceedings of the Institution of Mechanical Engineers, Part O: Journal of Risk and Reliability

publication icon
Oparaji U (2017) Robust artificial neural network for reliability and sensitivity analyses of complex non-linear systems. in Neural networks : the official journal of the International Neural Network Society

publication icon
Tolo S (2019) Robust on-line diagnosis tool for the early accident detection in nuclear power plants in Reliability Engineering & System Safety

Description We have developed efficient and reliable tools for on-line monitoring able to deal with uncertainty.
Not only we can predict the size of the LOCA but also have confidence associated with the prediction.
Exploitation Route The methodology developed is general and can be adopted to predict with confidence any signal of interest in different sectors. The methodology has been also integrated on the open source software OpenCossan
Sectors Aerospace, Defence and Marine,Chemicals,Creative Economy,Digital/Communication/Information Technologies (including Software),Energy,Manufacturing, including Industrial Biotechology,Transport

Description The research has allowed developing robust methodologies for on-line monitoring of safety-critical systems. Of one the main challenges were to provide to the operator and to the analyst some confidence in the methodology. To address this challenge we have developed robust neural networks supported by Bayesian statistics. The developed tool not only provides a reliable estimation of the state of the system but provides also some confidence bounds associated with the prediction. In addition, the developed tool allows dealing with noise data and conflicting information coming from sensors. The methodology has been further improved by Jonathan Sadeghi in his PhD thesis. The developed methods have been released as open-source software and available on the git repository of OpenCossan software (
First Year Of Impact 2017
Sector Digital/Communication/Information Technologies (including Software),Energy
Description FIS360 Innovation Consultant
Amount £25,000 (GBP)
Funding ID NNL/GC_551 
Organisation National Physical Laboratory 
Sector Academic/University
Country United Kingdom
Start 12/2021 
End 03/2022
Description Impact Acceleration Account
Amount £15,000 (GBP)
Organisation University of Liverpool 
Sector Academic/University
Country United Kingdom
Start 02/2019 
End 07/2019
Title Smartool 
Description Open source matlab toolbox for on-line monitoring and robust prediction 
Type Of Material Computer model/algorithm 
Year Produced 2018 
Provided To Others? Yes  
Impact Publication on Reliability Engineering & System Safety Volume 186, June 2019, Pages 110-119 
Description Case study for SMART project 
Organisation Bhabbha Atomic Research Centre
Country India 
Sector Public 
PI Contribution Based on the simulated accident data provided by BARC, a robust online diagnostic system for the nuclear reactor. The aim of the analysis is to identify the severity of the Loss Of Coolant Accident (LOCA) events on the basis of selected instrument signals: for this purpose, the main task of the implemented Artificial Neural Network is to recognize the pattern drawn by the input signals in time and to provide, on the basis of this information, the severity of the break in output. This is quantified in terms of break size, expressed in comparison with the double-ended rupture of the largest pipe in the reactor coolant system. For instance, a 200% break indicates the free discharge of the primary coolant from both the broken ends of the main pipe (this is generally considered the worst accident that can occur in a water circuit).
Collaborator Contribution The project partner simulated accident scenario of the primary heat transport system of a 220MWe pressurised heavy water reactor, whose design has a double containment with a vapour suppression pool. The main aim of the containment is to limit the release of radioactivity under normal and accident conditions, both at the ground level and through the stack. The accident scenario is a Loss Of Coolant Accident (LOCA) involving a double ended guillotine rupture of the reactor inlet header. In case of such accident, the vapour suppression pool is designed to limit the peak pressure and temperature in the containment, allowing the complete condensation of the incoming steam and limiting the leakage of fission products to the surrounding environment. In addition to this, several strategies (e.g. dissolving, trapping, entraining mechanisms) are in place to perform the removal of the fission products that reach the pool.
Impact Development of open source toolbox for on-line monitoring of critical systems
Start Year 2017
Description Development of online diagnostic system for nuclear power plants 
Organisation University of Portsmouth
Country United Kingdom 
Sector Academic/University 
PI Contribution Development of an opensource toolbox for robust online monitoring system
Collaborator Contribution Training of Artificial Neural Networks with different architectures
Impact Development of an opensource toolbox for robust online monitoring system
Start Year 2017
Title SMARTool 
Description All the methods developed in this research project have been implemented in to a open source MATLAB toolbox named SMARTool. The SMARTool contains procudures and scripts used to indentify and training ANN architectures adopting the Machine Learning Toolbox for MATLAB, while the pruning of the one hidden layer architecture has been performed adopting the NNSYSID toolbox. The NNSYSID is a toolbox for the identification of non-linear dynamic systems with artificial neural networks, often used as a benchmark, which implements several algorithms for ANN training and pruning. The SMARTool toolbox provides the adaptive Bayesian model averaging method, dataset organization and re-population methods (i.e. for linear and cubic spline interpolation as well as for Gaussian mixture sampling) and dedicated uncertainty quantification techniques, such as the error estimation for series association. In addition, the SMARTool can also access the advanced uncertainty quantification techniques provided by OpenCossan. 
Type Of Technology Software 
Year Produced 2018 
Open Source License? Yes  
Impact The open source toolbox is freely available and it can easily be integrated with other software 
Title Toolboox for OpenCossan 
Description OpenCOSSAN is a tool for uncertainty quantification and management. It represents the core of COSSAN software under continuous development at the Institute for Risk and Uncertainty,University of Liverpool, UK. All the algorithms and methods have been coded in a Matlab toolbox allowing numerical analysis, reliability analysis, simulation, sensitivity, optimization, robust design. OpenCossan is coded exploiting the object-oriented Matlab programming environment, where it is possible to define specialized solution sequences, which include reliability methods, sensitivity analysis, optimization strategies, surrogate models and parallel computing strategies. The computational framework is organized in packages. A package is a namespace for organizing classes and interfaces in a logical manner, which makes large software project OpenCossan easier to manage. A class describes a set of objects with common characteristics such as data structures and methods. Objects, that are instances of classes can be aggregated forming more complex objects and proving solutions for practical problem in a compact, organized and manageable format. The structure of the software allows for extensive modularity and efficient code re-utilization. Objects (instances of a class) can be aggregated forming more complex objects with methods providing solutions for practical problem in a compact, organized and manageable format. 
Type Of Technology Software 
Year Produced 2017 
Open Source License? Yes  
Impact Bayesian Belief Networks, more commonly known as Bayesian Networks, are a probabilistic graphical model based on the use of directed acyclic graphs, integrating graph theory with the robustness of Bayesian statistics. The graphical framework of such models consists of nodes, representing the variables of the problem of interest, connected to each other by edges, generally arrows, that depict the dependency link existing between two nodes. The main aim of the Bayesian Network approach is to factorize the probability of a complex event exploiting the knowledge regarding the dependencies existing among its sub-parts. In order to overcome the limitations associated with traditional Bayesian Networks, the integration of such approach with the imprecise probability theory has attracted increasing attention in the scientific community leading to the formulation and study of Credal Networks. Further efforts and research are strongly required in order to enhance the attractivness of Credal Networks outside the academic world and to ensure the reliability and efficiency of their performance in real-world applications. These aims represent the core of the Credal Networks toolbox developed within the OpenCossan framework: well known and novel methodologies are integrated in the software in order to provide the implementation, manipulation and analysis of Credal Networks.